Research On Radar Emitter Identification Technology Based On Deep Learning

Radar emitter identification is an important part of radar confrontation and plays a vital role in electronic warfare.With the development of science and technology,new radars and electronic warfare equipments are constantly emerging,and the modulation modes and parameters of signals become more complex and diverse.It is difficult to obtain satisfactory results based on radar signal recognition technology based on traditional signal characteristics.The application of artificial intelligence and signal intrapulse features to radar emitter identification has become an important research direction.Aiming at the problem that the recognition performance of existing radar emitter identification technology is significantly reduced under low SNR,this paper combines the time-frequency analysis method with the convolutional neural network(CNN)and the recurrent neural network(RNN)which have achieved excellent results in deep learning.Source identification technology.The research work in this paper is as follows:The common radar signal source signal models are discussed,and their time and frequency domain characteristics are simulated.The three time-frequency analysis methods,such as Morlet wavelet transform,Wigner-Ville distribution and smooth pseudo-Wigner-Ville distribution,are studied.And compare the performance of several methods through simulation.The common models in deep learning are analyzed.The structural functions,training algorithms and training processes of convolutional neural networks and cyclic neural networks are deeply studied.The problems in training are analyzed and solved.A radar emitter identification method based on optimized CNN is proposed and the process of identification is given.The time-frequency diagram of the radar source signal is obtained by the time-frequency analysis method based on Morlet wavelet transform.The preprocessing method for time-frequency diagram is discussed.The structure of the convolutional neural network is optimized by introducing the dropout layer,the Softmax layer,and the Tensorflow framework.The simulation results of radar radiation source signals with different modulation modes and different modulation parameters are presented.The results show that the proposed method not only has excellent performance under high SNR conditions,but also has good recognition effect under low signal and noise.Even if the signal-to-noise ratio is as low as-14 d B,under certain simulation conditions,the radar source signal of five different modulation modes and the radar modulation source signals of different modulation parameters of the same modulation mode still have a very excellent recognition effect.The radar radiation source identification method based on Recurrent Neural Network is studied and the identification process is given.The time-frequency diagram generated by the radar radiation source signal is segmented according to the time axis,and the time-frequency sequence is obtained as the input feature vector.For cyclic neural networks,memory cells are added in their hidden layers to form a long-term and short-term memory network LSTM.The simulation proves that the method shortens the simulation time under the premise of little impact on performance.When the signal-to-noise ratio is as low as-8d B,under certain simulation conditions,the radar radiation source signals of five different modulation modes and the radar modulation source signals of different modulation parameters of the same modulation mode still have satisfactory recognition effects.Simulation experiments show that the time-frequency analysis method has little effect on the recognition rate and other performance,but the input image size,learning rate,and number of hidden units will affect the recognition performance.